DEVELOPER DOCUMENTATION

Quickstart

From an idea to your first completion. One key, one compatible API.

GPU Router supports text Chat Completions, function tools, and structured JSON on routes that advertise those features. Selected models also support image inputs; see the coding-agent image guide for supported shapes and limits. Audio inputs are unsupported.

1. Create a key

Sign in, open API Keys, and create a key. Copy the key when it appears; it is shown once. Keep it in a server environment variable.

New accounts automatically receive $1 in free credits. No payment is required to start testing; add more credits in Billing when needed.

2. Choose an available model

Browse models available on your router. Use the exact gateway model ID. A public market listing does not guarantee an executable route on your network.

3. Make a request

Set your client’s base URL to https://gpurouter.io/v1. Paid requests use the funding source selected for your key; the gateway verifies your available balance and key allowance before dispatch.

These examples pin the selected billing source. Earned-only requests do not charge prepaid funds. Deferred batches with a billing assertion are not supported yet.

curl 'https://gpurouter.io/v1/chat/completions' \
  -H "Authorization: Bearer $GPU_ROUTER_API_KEY" \
  -H 'Content-Type: application/json' \
  -H 'x-gpu-billing: {"version":1,"source":"prepaid"}' \
  --data-binary '{
  "model": "your-model-id",
  "messages": [
    {
      "role": "user",
      "content": "Hello!"
    }
  ],
  "max_tokens": 256
}'

Need zero data retention? Add "zdr": true to every request. Follow the ZDR guide for eligible models and SDK examples.

4. Inspect the result

Connecting an editor? Follow the coding-agent compatibility guides.

Open Activity for token counts, cost, outcome, and latency. Or try a model in the side playground.

Ask the docs assistant

Get AI help with integration, pricing, routing, batches, latency, and privacy, with links to the relevant documentation. The assistant can explain and suggest code; it has no account access and cannot execute requests or change settings.

Try three guest questions per browser session each UTC day. A ten-second cooldown and shared usage limits also apply. Questions go to the configured AI provider; don’t include secrets, and check the source links for accuracy.